That is essentially correct, but be careful with the direction part. For example, imagine I am trading equities. I might have say a bond position with a negative correlation of -.85 to the equities position. So even though I might be long both, they are some sort of pair trade from a comovement perspective. Think of all the classical cointegrated trades, the crack spread, the Nob/Schatz/Bobl, TB/Eurodollar butterflies, etc etc. What I do differently is that I don't trade these instruments because I want two things: 1) Very strong underlying cointegration of the symbols. Not just statistical, but logical. That is satisfied by classical trades. 2) But, I also don't want them well known. This is key, imo. The idea is to get as close to statistical arbitrage as possible. The ultimate would be say triangular FOREX trades. But that takes huge resources. You could trade classical cointegrated trades and tons of traders make a good living doing it, but then you should go to an institution where you can have $10,000,000 notional position on. Otherwise, as a retail trade with $100,000 account, even with portfolio margining, you will probably starve. But, you won't go broke either! Another caution, I don't trade equity pairs. I think you need really deep pockets for that. Merger Arb or say two symbols on the same underlying, e.g., BRK.A BRK.B are the only thing that a retail trader should trade. The risks are enormous otherwise imo. Remember, the key is to be able to scale to huge size repeatedly, and hope the once in a lifetime event doesn't get you (a repeat of 2008-2009 would probably blow me out, or I might make a fortune, but that is too high an ulcer trade for my taste). Then let the logic of the position play out on whatever time frame you have done the trade on. In order to do that you have to have odds of blowing out at 1:50000 (one in fifty thousand) or better. In other words, you are more likely to die on an airplane flight.
Nitro, How little known a synthetic spread do you feel is reasonable.? Software can scale volatility or detrend a spread so that it is MR. Does it make sense to do this? In another case , Matching 3 or 4 assets can lead to a MReverting spread but who is to say how much sense , for example, gold,yen and treasuries viewed as a single position makes at the time. Also, say you do see something such as a level respected a few times recently. An example may be a spread of Brent and Emerging markets. I have seen this turn like clockwork yet i cannot reason this event by the fumdamentals and PA of the seperate legs and attribute it to a coincidence of the arithmetic. Maybe it is a tradeoff. That there is a lot of the above going on that belongs the bucket, but , maybe one or two that dont. Telling the difference seems to be the rub. Great thread.
As long as the spread is logical, it makes no difference who trades it. Then it is a matter of modeling it correctly, and in my case coding the model. When you have lots of instruments it is wise to do PCA. Don't just look at charts. Don't be afraid to trade what you see. Logic has to make sense to me, but maybe for more risk and greater reward maybe statistics and probability is enough. Trade small. I am never afraid to blow out ten small accounts. If the eleventh works, I make it all back many times over. The hard part is putting on size.
@nitro I assume when you say logical you mean fundamental? i.e. the assets are bound by some fundamental reason. It is interesting that you avoid equities, I don't exclude any particular asset classes. I base selections on fundamental reasons, for example, assets that share a common underlying bias where that underlying may be a commodity, interest rate, or semiconductor prices, etc. I take a bunch of assets with a fundamental bias and rank them according to strength of reversion at various time frames. The beauty of this game is there are literally an unlimited number of markets you can create. I'm a professional software developer and use my own software models to do this. I trade small and use pdf to ensure the odds are skewed in my favour. The next trick is to expand the selections and utilise assets with varying currencies, do you do this? I'm not sure of the best way to do this, should I convert the prices 'on-the-fly' to a common currency, or should I hedge the currency risk for the individual legs ?
Sort of, well, mostly. I don't want to prejudice it too much. Let me put it this way, sometimes it is not obvious at all what the connection is, but you swear you see it in the comovement. It is truly enlightening when you discover the connection. So, yes, but stay adaptable and open minded to connections that don't make sense at first blush. Well, I just think that per dollar invested/traded and risk taken, equities are one of the worst trading venues. Also, I hate things that can be manipulated by endless people talking about it. In other words, I prefer something with a high signal to noise ratio (high signal low noise). That said, there are some good equities trading strategies. I have done it in the past. It is a reasonable way to trade, but I find it more directional than I like. The way to reduce risk in this case is strictly statistical, and scaling this sort of model to big size is not my taste. Doesn't mean it can't work. Hedging currency risk is not easy. Lots of businesses have to engage in it, and most of them get it wrong. Let me put it this way, if you could get the currency part correct, I wouldn't even bother with the rest because currencies are massively scalable. There are so many moving parts to currency hedging that it is probably close to impossible unless you are an institution. Check the news that moves these things non-trivially every day: http://www.forexfactory.com/ And every day it looks like that. I don't have an answer.
@nitro Thank you for the information. What is the typical half-life / trade time for your spreads? I know this is a how long is a piece of string question, but a ball park figure would be interesting. I'd like to know if you are intra-day spread trading or more longer-term (a few days)?
It depends! I trade around my core position, in and out as I see patterns in the individual instruments. I find that I can increase returns and lower risk that way. The core position is often two weeks. The "Epicycles" are one or two days long. My current position had an Adverse Maximum Excursion on the Paris attacks. But systems said hold and it snapped back violently. There are so many cogs and wheels in my system(s) that I feel like what I am doing is closer to adding Epicycles than "Quantum Field Theory" (QFT). Interestingly, the holistic system is discovered with "QFT", but in order to make it work I have to touch one foot to the ground and trade the realities of the world. You can't just do Physics. I wonder how a machine learning approach could ever even come close to discovering any of this. You might as well have a monkey typing randomly on a keyboard trying to recreate Shakespeare.
Here is a typical complication. This is the probability distribution of an instrument at high frequency (almost a half million quotes) over just a few days. It doesn't matter what it is and ignore the scale. How do you fit a model to this? Notice it is nowhere near Normal (news really messes everything). What approach would you take?
Damn it Jim I'm a software engineer not a mathematician! Having said that, the distribution does have tails, even though it is not Gaussian like. The model could take positions at the extremes (say 95%) and wait for reversion?
Possible, but can you see a way to do S/R trading using this information? What about regime shifts? A market making algorithm could instead of using a single pdf, it could use a multinomial one. Still, the challenges are maddening.